🌟 Data Integration and Cell Type Classification of scRNA-seq Pancreas Dataset 🌟

🚀 Introduction

Welcome to our group 5 project repository! 🧬 This project is all about charting new territories in the pancreas using single-cell RNA sequencing (scRNA-seq) data. We're on a mission to conquer the challenges of batch effect correction and data high-dimensionality in scRNA-seq with some seriously advanced computational firepower! 💻

🛠 Methods

We've deployed three main computational wizards:

1. Unsupervised Clustering (MNN + KMeans): 🔍 Identifying distinct cell types.
2. GLM with Lasso Penalty: 📐 Enhancing feature selection.
3. Multi-Task Neural Network: 🧠 A top-notch approach for cell type classification.

🌌 Results

Behold the power of our methods, especially the multi-task neural network, in classifying cell types with astronomical accuracy! 🌠

• 📊 UMAP Visualization:

  

  

  

• 📈 ARI and NMI Index:

Model	ARI	NMI
Unsupervised Clustering MNN + kmeans	0.5717	0.6139
GLM with Lasso penalty	0.9502	0.9146
Transfer Learning and Multi-task Neural Network	1.0000	1.0000

👥 Collaboration

A big shout-out to our cosmic crew 🚀, who contributed to data processing, model development, and analysis. Every part of this project has been crafted with utmost care and precision. Introduction and data processing was contributed by Xintong. GLM with Lasso penalty was contributed by Bulun, MNN + kmeans was contributed by Kexin, Multi-task Neural Network was contributed by Xintong. The final report was collaboratively drafted by all team members.

📁 Repository Structure

• DeepNeuralNetwork/: 🤖 Jupyter notebooks for transfer-learning and multilevel classification implemented with pytorch framework.
• MNN/: 🧼 Unsupervised learning via Mutual Nearest Neighbors, acclerated with Rcpp
• SoftMaxLasso/: 🧼 Generalized Logistic regression model with L1 regularization optimized by cpp-acclerated FISTA algorithm
• Data/: 📊 Original H5 Data preprocessed_adata.h5ad , PCA dimension reducted adata_pca.csv, and gene expression data x_matrix. Large files have been prtitioned into several zip files.

Thank you for visiting our project! We welcome your contributions and feedback to make this project even more !